1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
5 /* Copyright (c) University of Cambridge 1995 - 2015 */
6 /* Copyright (c) The Exim Maintainers 2020 */
7 /* See the file NOTICE for conditions of use and distribution. */
9 /* A set of functions to search databases in various formats. An open
10 database is represented by a void * value which is returned from a lookup-
11 specific "open" function. These are now all held in individual modules in the
12 lookups subdirectory and the functions here form a generic interface.
14 Caching is used to improve performance. Open files are cached until a tidyup
15 function is called, and for each file the result of the last lookup is cached.
16 However, if too many files are opened, some of those that are not in use have
17 to be closed. Those open items that use real files are kept on a LRU chain to
20 All the data is held in permanent store so as to be independent of the stacking
21 pool that is reset from time to time. In fact, we use malloc'd store so that it
22 can be freed when the caches are tidied up. It isn't actually clear whether
23 this is a benefit or not, to be honest. */
28 /* Tree in which to cache open files until tidyup called. */
30 static tree_node *search_tree = NULL;
32 /* Two-way chain of open databases that use real files. This is maintained in
33 recently-used order for the purposes of closing the least recently used when
34 too many files are open. */
36 static tree_node *open_top = NULL;
37 static tree_node *open_bot = NULL;
39 /* Count of open databases that use real files */
41 static int open_filecount = 0;
43 /* Allow us to reset store used for lookups and lookup caching */
45 static rmark search_reset_point = NULL;
49 /*************************************************
50 * Validate a plain lookup type name *
51 *************************************************/
53 /* Only those names that are recognized and whose code is included in the
54 binary give an OK response. Use a binary chop search now that the list has got
58 name lookup type name - not necessarily zero terminated (e.g. dbm*)
59 len length of the name
61 Returns: +ve => valid lookup name; value is offset in lookup_list
62 -ve => invalid name; message in search_error_message.
66 search_findtype(const uschar *name, int len)
69 int top = lookup_list_count;
72 int mid = (top + bot)/2;
73 int c = Ustrncmp(name, lookup_list[mid]->name, len);
75 /* If c == 0 we have matched the incoming name with the start of the search
76 type name. However, some search types are substrings of others (e.g. nis and
77 nisplus) so we need to check that the lengths are the same. The length of the
78 type name cannot be shorter (else c would not be 0); if it is not equal it
79 must be longer, and in that case, the incoming name comes before the name we
80 are testing. By leaving c == 0 when the lengths are different, and doing a
81 > 0 test below, this all falls out correctly. */
83 if (c == 0 && Ustrlen(lookup_list[mid]->name) == len)
85 if (lookup_list[mid]->find != NULL) return mid;
86 search_error_message = string_sprintf("lookup type \"%.*s\" is not "
87 "available (not in the binary - check buildtime LOOKUP configuration)",
92 if (c > 0) bot = mid + 1; else top = mid;
95 search_error_message = string_sprintf("unknown lookup type \"%.*s\"",len,name);
101 /*************************************************
102 * Validate a full lookup type name *
103 *************************************************/
105 /* This function recognizes the "partial-" prefix and also terminating * and *@
109 name the full lookup type name
110 ptypeptr where to put the partial type
111 after subtraction of 1024 or 2048:
112 negative => no partial matching
113 non-negative => minimum number of non-wild components
114 ptypeaff where to put a pointer to the affix
115 the affix is within name if supplied therein
116 otherwise it's a literal string
117 afflen the length of the affix
118 starflags where to put the SEARCH_STAR and SEARCH_STARAT flags
119 opts where to put the options
121 Returns: +ve => valid lookup name; value is offset in lookup_list
122 -ve => invalid name; message in search_error_message.
126 search_findtype_partial(const uschar *name, int *ptypeptr, const uschar **ptypeaff,
127 int *afflen, int *starflags, const uschar ** opts)
131 const uschar *ss = name;
137 /* Check for a partial matching type. It must start with "partial", optionally
138 followed by a sequence of digits. If this is followed by "-", the affix is the
139 default "*." string. Otherwise we expect an affix in parentheses. Affixes are a
140 limited number of characters, not including parens. */
142 if (Ustrncmp(name, "partial", 7) == 0)
148 while (isdigit(*ss)) pv = pv*10 + *ss++ - '0';
150 else pv = 2; /* Default number of wild components */
155 while (ispunct(*ss) && *ss != ')') ss++;
156 if (*ss != ')') goto BAD_TYPE;
157 *afflen = ss++ - *ptypeaff;
159 else if (*ss++ == '-')
167 search_error_message = string_sprintf("format error in lookup type \"%s\"",
173 /* Now we are left with a lookup name, possibly followed by * or *@,
174 and then by options starting with a "," */
177 if ((t = Ustrchr(ss, '*')))
180 *starflags |= (t[1] == '@' ? SEARCH_STARAT : SEARCH_STAR);
185 if ((t = Ustrchr(t, ',')))
188 if (l < len) len = l;
189 *opts = string_copy(t+1);
194 /* Check for the individual search type. Only those that are actually in the
195 binary are valid. For query-style types, "partial" and default types are
198 stype = search_findtype(ss, len);
199 if (stype >= 0 && mac_islookup(stype, lookup_querystyle))
203 search_error_message = string_sprintf("\"partial\" is not permitted "
204 "for lookup type \"%s\"", ss);
207 if ((*starflags & (SEARCH_STAR|SEARCH_STARAT)) != 0)
209 search_error_message = string_sprintf("defaults using \"*\" or \"*@\" are "
210 "not permitted for lookup type \"%s\"", ss);
221 /*************************************************
222 * Release cached resources *
223 *************************************************/
225 /* When search_open is called it caches the "file" that it opens in
226 search_tree. The name of the tree node is a concatenation of the search type
227 with the file name. For query-style lookups, the file name is empty. Real files
228 are normally closed only when this tidyup routine is called, typically at the
229 end of sections of code where a number of lookups might occur. However, if too
230 many files are open simultaneously, some get closed beforehand. They can't be
231 removed from the tree. There is also a general tidyup function which is called
232 for the lookup driver, if it exists.
234 First, there is an internal, recursive subroutine.
236 Argument: a pointer to a search_openfile tree node
241 tidyup_subtree(tree_node *t)
243 search_cache * c = (search_cache *)(t->data.ptr);
244 if (t->left) tidyup_subtree(t->left);
245 if (t->right) tidyup_subtree(t->right);
246 if (c && c->handle && lookup_list[c->search_type]->close)
247 lookup_list[c->search_type]->close(c->handle);
251 /* The external entry point
260 int old_pool = store_pool;
262 DEBUG(D_lookup) debug_printf_indent("search_tidyup called\n");
264 /* Close individually each cached open file. */
266 store_pool = POOL_SEARCH;
269 tidyup_subtree(search_tree);
272 open_top = open_bot = NULL;
275 /* Call the general tidyup entry for any drivers that have one. */
277 for (int i = 0; i < lookup_list_count; i++) if (lookup_list[i]->tidy)
278 (lookup_list[i]->tidy)();
280 if (search_reset_point) search_reset_point = store_reset(search_reset_point);
281 store_pool = old_pool;
287 /*************************************************
288 * Open search database *
289 *************************************************/
291 /* A mode, and lists of owners and groups, are passed over for checking in
292 the cases where the database is one or more files. Return NULL, with a message
293 pointed to by message, in cases of error.
295 For search types that use a file or files, check up on the mode after
296 opening. It is tempting to do a stat before opening the file, and use it as
297 an existence check. However, doing that opens a small security loophole in
298 that the status could be changed before the file is opened. Can't quite see
299 what problems this might lead to, but you can't be too careful where security
300 is concerned. Fstat() on an open file can normally be expected to succeed,
301 but there are some NFS states where it does not.
303 There are two styles of query: (1) in the "single-key+file" style, a single
304 key string and a file name are given, for example, for linear searches, DBM
305 files, or for NIS. (2) In the "query" style, no "filename" is given; instead
306 just a single query string is passed. This applies to multiple-key lookup
309 Before opening, scan the tree of cached files to see if this file is already
310 open for the correct search type. If so, return the saved handle. If not, put
311 the handle in the tree for possible subsequent use. See search_tidyup above for
312 closing all the cached files.
314 A count of open databases which use real files is maintained, and if this
315 gets too large, we have to close a cached file. Its entry remains in the tree,
316 but is marked closed.
319 filename the name of the file for single-key+file style lookups,
320 NULL for query-style lookups
321 search_type the type of search required
322 modemask if a real single file is used, this specifies mode bits that
323 must not be set; otherwise it is ignored
324 owners if a real single file is used, this specifies the possible
325 owners of the file; otherwise it is ignored
326 owngroups if a real single file is used, this specifies the possible
327 group owners of the file; otherwise it is ignored
329 Returns: an identifying handle for the open database;
330 this is the pointer to the tree block in the
331 cache of open files; return NULL on open failure, with
332 a message in search_error_message
336 search_open(const uschar * filename, int search_type, int modemask,
337 uid_t * owners, gid_t * owngroups)
342 lookup_info *lk = lookup_list[search_type];
343 uschar keybuffer[256];
344 int old_pool = store_pool;
346 if (filename && is_tainted(filename))
348 log_write(0, LOG_MAIN|LOG_PANIC,
349 "Tainted filename for search: '%s'", filename);
353 /* Change to the search store pool and remember our reset point */
355 store_pool = POOL_SEARCH;
356 if (!search_reset_point) search_reset_point = store_mark();
358 DEBUG(D_lookup) debug_printf_indent("search_open: %s \"%s\"\n", lk->name,
359 filename ? filename : US"NULL");
361 /* See if we already have this open for this type of search, and if so,
362 pass back the tree block as the handle. The key for the tree node is the search
363 type plus '0' concatenated with the file name. There may be entries in the tree
364 with closed files if a lot of files have been opened. */
366 sprintf(CS keybuffer, "%c%.254s", search_type + '0',
367 filename ? filename : US"");
369 if ((t = tree_search(search_tree, keybuffer)))
371 if ((c = (search_cache *)t->data.ptr)->handle)
373 DEBUG(D_lookup) debug_printf_indent(" cached open\n");
374 store_pool = old_pool;
377 DEBUG(D_lookup) debug_printf_indent(" cached closed\n");
380 /* Otherwise, we need to open the file or database - each search type has its
381 own code, which is now split off into separately compiled modules. Before doing
382 this, if the search type is one that uses real files, check on the number that
383 we are holding open in the cache. If the limit is reached, close the least
384 recently used one. */
386 if (lk->type == lookup_absfile && open_filecount >= lookup_open_max)
388 log_write(0, LOG_MAIN|LOG_PANIC, "too many lookups open, but can't find "
392 search_cache *c = (search_cache *)(open_bot->data.ptr);
393 DEBUG(D_lookup) debug_printf_indent("Too many lookup files open\n closing %s\n",
395 if ((open_bot = c->up))
396 ((search_cache *)(open_bot->data.ptr))->down = NULL;
399 ((lookup_list[c->search_type])->close)(c->handle);
404 /* If opening is successful, call the file-checking function if there is one,
405 and if all is still well, enter the open database into the tree. */
407 if (!(handle = (lk->open)(filename, &search_error_message)))
409 store_pool = old_pool;
414 && !lk->check(handle, filename, modemask, owners, owngroups,
415 &search_error_message))
418 store_pool = old_pool;
422 /* If this is a search type that uses real files, keep count. */
424 if (lk->type == lookup_absfile) open_filecount++;
426 /* If we found a previously opened entry in the tree, re-use it; otherwise
427 insert a new entry. On re-use, leave any cached lookup data and the lookup
432 t = store_get(sizeof(tree_node) + Ustrlen(keybuffer), FALSE);
433 t->data.ptr = c = store_get(sizeof(search_cache), FALSE);
434 c->item_cache = NULL;
435 Ustrcpy(t->name, keybuffer);
436 tree_insertnode(&search_tree, t);
438 else c = t->data.ptr;
441 c->search_type = search_type;
442 c->up = c->down = NULL;
444 store_pool = old_pool;
452 /*************************************************
453 * Internal function: Find one item in database *
454 *************************************************/
456 /* The answer is always put into dynamic store. The last lookup for each handle
460 handle the handle from search_open; points to tree node
461 filename the filename that was handed to search_open, or
462 NULL for query-style searches
463 keystring the keystring for single-key+file lookups, or
464 the querystring for query-style lookups
465 opts type-specific options
467 Returns: a pointer to a dynamic string containing the answer,
468 or NULL if the query failed or was deferred; in the
469 latter case, search_find_defer is set TRUE; after an unusual
470 failure, there may be a message in search_error_message.
474 internal_search_find(void * handle, const uschar * filename, uschar * keystring,
477 tree_node * t = (tree_node *)handle;
478 search_cache * c = (search_cache *)(t->data.ptr);
479 expiring_data * e = NULL; /* compiler quietening */
480 uschar * data = NULL;
481 int search_type = t->name[0] - '0';
482 int old_pool = store_pool;
484 /* Lookups that return DEFER may not always set an error message. So that
485 the callers don't have to test for NULL, set an empty string. */
487 search_error_message = US"";
488 f.search_find_defer = FALSE;
490 DEBUG(D_lookup) debug_printf_indent("internal_search_find: file=\"%s\"\n "
491 "type=%s key=\"%s\" opts=%s%s%s\n", filename,
492 lookup_list[search_type]->name, keystring,
493 opts ? "\"" : "", opts, opts ? "\"" : "");
495 /* Insurance. If the keystring is empty, just fail. */
497 if (keystring[0] == 0) return NULL;
499 /* Use the special store pool for search data */
501 store_pool = POOL_SEARCH;
503 /* Look up the data for the key, unless it is already in the cache for this
504 file. No need to check c->item_cache for NULL, tree_search will do so. */
506 if ( (t = tree_search(c->item_cache, keystring))
507 && (!(e = t->data.ptr)->expiry || e->expiry > time(NULL))
508 && (!opts && !e->opts || opts && e->opts && Ustrcmp(opts, e->opts) == 0)
510 { /* Data was in the cache already; set the pointer from the tree node */
512 DEBUG(D_lookup) debug_printf_indent("cached data used for lookup of %s%s%s\n",
514 filename ? US"\n in " : US"", filename ? filename : US"");
518 uint do_cache = UINT_MAX;
519 int keylength = Ustrlen(keystring);
524 debug_printf_indent("cached data found but %s; ",
525 e->expiry && e->expiry <= time(NULL) ? "out-of-date" : "wrong opts");
526 debug_printf_indent("%s lookup required for %s%s%s\n",
527 filename ? US"file" : US"database",
529 filename ? US"\n in " : US"", filename ? filename : US"");
532 /* Call the code for the different kinds of search. DEFER is handled
533 like FAIL, except that search_find_defer is set so the caller can
534 distinguish if necessary. */
536 if (lookup_list[search_type]->find(c->handle, filename, keystring, keylength,
537 &data, &search_error_message, &do_cache, opts) == DEFER)
538 f.search_find_defer = TRUE;
540 /* A record that has been found is now in data, which is either NULL
541 or points to a bit of dynamic store. Cache the result of the lookup if
542 caching is permitted. Lookups can disable caching, when they did something
543 that changes their data. The mysql and pgsql lookups do this when an
544 UPDATE/INSERT query was executed. */
548 if (!t) /* No existing entry. Create new one. */
550 int len = keylength + 1;
551 e = store_get(sizeof(expiring_data) + sizeof(tree_node) + len, is_tainted(keystring));
552 t = (tree_node *)(e+1);
553 memcpy(t->name, keystring, len);
555 tree_insertnode(&c->item_cache, t);
557 /* Else previous, out-of-date cache entry. Update with the */
558 /* new result and forget the old one */
559 e->expiry = do_cache == UINT_MAX ? 0 : time(NULL)+do_cache;
560 e->opts = opts ? string_copy(opts) : NULL;
564 /* If caching was disabled, empty the cache tree. We just set the cache
565 pointer to NULL here, because we cannot release the store at this stage. */
569 DEBUG(D_lookup) debug_printf_indent("lookup forced cache cleanup\n");
570 c->item_cache = NULL; /* forget all lookups on this connection */
577 debug_printf_indent("lookup yielded: %s\n", data);
578 else if (f.search_find_defer)
579 debug_printf_indent("lookup deferred: %s\n", search_error_message);
580 else debug_printf_indent("lookup failed\n");
583 /* Return it in new dynamic store in the regular pool */
585 store_pool = old_pool;
586 return data ? string_copy(data) : NULL;
592 /*************************************************
593 * Find one item in database, possibly wildcarded *
594 *************************************************/
596 /* This function calls the internal function above; once only if there
597 is no partial matching, but repeatedly when partial matching is requested.
600 handle the handle from search_open
601 filename the filename that was handed to search_open, or
602 NULL for query-style searches
603 keystring the keystring for single-key+file lookups, or
604 the querystring for query-style lookups
605 partial -1 means no partial matching;
606 otherwise it's the minimum number of components;
607 affix the affix string for partial matching
608 affixlen the length of the affix string
609 starflags SEARCH_STAR and SEARCH_STARAT flags
610 expand_setup pointer to offset for setting up expansion strings;
612 opts type-specific options
614 Returns: a pointer to a dynamic string containing the answer,
615 or NULL if the query failed or was deferred; in the
616 latter case, search_find_defer is set TRUE
620 search_find(void * handle, const uschar * filename, uschar * keystring,
621 int partial, const uschar * affix, int affixlen, int starflags,
622 int * expand_setup, const uschar * opts)
624 tree_node *t = (tree_node *)handle;
625 BOOL set_null_wild = FALSE;
630 if (partial < 0) affixlen = 99; /* So that "NULL" prints */
631 debug_printf_indent("search_find: file=\"%s\"\n key=\"%s\" "
632 "partial=%d affix=%.*s starflags=%x opts=%s%s%s\n",
633 filename ? filename : US"NULL",
634 keystring, partial, affixlen, affix, starflags,
635 opts ? "\"" : "", opts, opts ? "\"" : "");
639 /* Arrange to put this database at the top of the LRU chain if it is a type
640 that opens real files. */
642 if ( open_top != (tree_node *)handle
643 && lookup_list[t->name[0]-'0']->type == lookup_absfile)
645 search_cache *c = (search_cache *)(t->data.ptr);
646 tree_node *up = c->up;
647 tree_node *down = c->down;
649 /* Cut it out of the list. A newly opened file will a NULL up pointer.
650 Otherwise there will be a non-NULL up pointer, since we checked above that
651 this block isn't already at the top of the list. */
655 ((search_cache *)(up->data.ptr))->down = down;
657 ((search_cache *)(down->data.ptr))->up = up;
662 /* Now put it at the head of the list. */
666 if (!open_top) open_bot = t;
667 else ((search_cache *)(open_top->data.ptr))->up = t;
673 debug_printf_indent("LRU list:\n");
674 for (tree_node *t = open_top; t; )
676 search_cache *c = (search_cache *)(t->data.ptr);
677 debug_printf_indent(" %s\n", t->name);
678 if (t == open_bot) debug_printf_indent(" End\n");
683 /* First of all, try to match the key string verbatim. If matched a complete
684 entry but could have been partial, flag to set up variables. */
686 yield = internal_search_find(handle, filename, keystring, opts);
687 if (f.search_find_defer) return NULL;
689 if (yield) { if (partial >= 0) set_null_wild = TRUE; }
691 /* Not matched a complete entry; handle partial lookups, but only if the full
692 search didn't defer. Don't use string_sprintf() to construct the initial key,
693 just in case the original key is too long for the string_sprintf() buffer (it
694 *has* happened!). The case of a zero-length affix has to be treated specially.
697 else if (partial >= 0)
699 int len = Ustrlen(keystring);
702 /* Try with the affix on the front, except for a zero-length affix */
704 if (affixlen == 0) keystring2 = keystring; else
706 keystring2 = store_get(len + affixlen + 1,
707 is_tainted(keystring) || is_tainted(affix));
708 Ustrncpy(keystring2, affix, affixlen);
709 Ustrcpy(keystring2 + affixlen, keystring);
710 DEBUG(D_lookup) debug_printf_indent("trying partial match %s\n", keystring2);
711 yield = internal_search_find(handle, filename, keystring2, opts);
712 if (f.search_find_defer) return NULL;
715 /* The key in its entirety did not match a wild entry; try chopping off
716 leading components. */
721 uschar *keystring3 = keystring2 + affixlen;
722 uschar *s = keystring3;
723 while (*s != 0) if (*s++ == '.') dotcount++;
725 while (dotcount-- >= partial)
727 while (*keystring3 != 0 && *keystring3 != '.') keystring3++;
729 /* If we get right to the end of the string (which will be the last time
730 through this loop), we've failed if the affix is null. Otherwise do one
731 last lookup for the affix itself, but if it is longer than 1 character,
732 remove the last character if it is ".". */
734 if (*keystring3 == 0)
736 if (affixlen < 1) break;
737 if (affixlen > 1 && affix[affixlen-1] == '.') affixlen--;
738 Ustrncpy(keystring2, affix, affixlen);
739 keystring2[affixlen] = 0;
740 keystring3 = keystring2;
744 keystring3 -= affixlen - 1;
745 if (affixlen > 0) Ustrncpy(keystring3, affix, affixlen);
748 DEBUG(D_lookup) debug_printf_indent("trying partial match %s\n", keystring3);
749 yield = internal_search_find(handle, filename, keystring3, opts);
750 if (f.search_find_defer) return NULL;
753 /* First variable is the wild part; second is the fixed part. Take care
754 to get it right when keystring3 is just "*". */
756 if (expand_setup && *expand_setup >= 0)
758 int fixedlength = Ustrlen(keystring3) - affixlen;
759 int wildlength = Ustrlen(keystring) - fixedlength - 1;
761 expand_nstring[*expand_setup] = keystring;
762 expand_nlength[*expand_setup] = wildlength;
764 expand_nstring[*expand_setup] = keystring + wildlength + 1;
765 expand_nlength[*expand_setup] = (fixedlength < 0)? 0 : fixedlength;
769 keystring3 += affixlen;
773 else set_null_wild = TRUE; /* Matched a wild entry without any wild part */
776 /* If nothing has been matched, but the option to look for "*@" is set, try
777 replacing everything to the left of @ by *. After a match, the wild part
778 is set to the string to the left of the @. */
780 if (!yield && starflags & SEARCH_STARAT)
782 uschar *atat = Ustrrchr(keystring, '@');
783 if (atat != NULL && atat > keystring)
786 savechar = *(--atat);
789 DEBUG(D_lookup) debug_printf_indent("trying default match %s\n", atat);
790 yield = internal_search_find(handle, filename, atat, opts);
792 if (f.search_find_defer) return NULL;
794 if (yield && expand_setup && *expand_setup >= 0)
797 expand_nstring[*expand_setup] = keystring;
798 expand_nlength[*expand_setup] = atat - keystring + 1;
800 expand_nstring[*expand_setup] = keystring;
801 expand_nlength[*expand_setup] = 0;
806 /* If we still haven't matched anything, and the option to look for "*" is set,
807 try that. If we do match, the first variable (the wild part) is the whole key,
808 and the second is empty. */
810 if (!yield && starflags & (SEARCH_STAR|SEARCH_STARAT))
812 DEBUG(D_lookup) debug_printf_indent("trying to match *\n");
813 yield = internal_search_find(handle, filename, US"*", opts);
814 if (yield && expand_setup && *expand_setup >= 0)
817 expand_nstring[*expand_setup] = keystring;
818 expand_nlength[*expand_setup] = Ustrlen(keystring);
820 expand_nstring[*expand_setup] = keystring;
821 expand_nlength[*expand_setup] = 0;
825 /* If this was a potentially partial lookup, and we matched either a
826 complete non-wild domain entry, or we matched a wild-carded entry without
827 chopping off any of the domain components, set up the expansion variables
828 (if required) so that the first one is empty, and the second one is the
829 fixed part of the domain. The set_null_wild flag is set only when yield is not
832 if (set_null_wild && expand_setup && *expand_setup >= 0)
835 expand_nstring[*expand_setup] = keystring;
836 expand_nlength[*expand_setup] = 0;
838 expand_nstring[*expand_setup] = keystring;
839 expand_nlength[*expand_setup] = Ustrlen(keystring);
845 /* End of search.c */